RESEARCH DOSSIER
Production Possibility Frontiers and Technological Tradeoffs
Abstract. Production possibility frontiers characterize the maximal combinations of outputs attainable under specified resources, technologies, institutional conditions, and time horizons. This monograph develops the economic meaning of those frontiers, derives their marginal transformation rates, investigates the conditions under which prices can decentralize productive choices, and demonstrates the consequences of indivisibility, joint production, technological bias, uncertainty, and intertemporal investment. The historical investigation connects classical price theory, general equilibrium, growth accounting, endogenous innovation, and frontier measurement. Formal derivations and explicitly hypothetical counterexamples distinguish resource reallocation, efficiency improvement, technological change, and changes in constraints. Comparative applications to agriculture, energy, manufacturing, trade, and public services show how measured opportunities depend on units, boundaries, available complements, and counterfactual accessibility. The analysis establishes conditions for interpreting a shift in the attainable set and explains why output growth, a reduction in unit cost, or movement toward a statistical envelope cannot independently identify technical progress. The contribution is a structured analytical synthesis, with reproducible illustrative calculations and a source-critical empirical design, rather than a new named theory or a causal estimate from newly collected data.
Keywords: production possibility frontier; production set; marginal rate of transformation; opportunity cost; nonconvex technology; joint production; technical change; productive efficiency; productivity measurement; economic tradeoffs; shadow prices; environmental constraints; comparative advantage
Feasibility and the economic problem
The object of investigation
An economy organizes finite resources into outputs through a collection of feasible activities. The production possibility frontier describes the maximal output configurations attainable within that collection. Its economic significance arises from the coexistence of alternative uses for productive resources. A resource committed to one activity may lose the capacity to support another, while improvements in organization can change the combinations that are jointly attainable.
The frontier is conditional. A farm, factory, municipality, national economy, or group of countries can each be represented by a feasible set, provided that the relevant period, resource endowments, technological practices, and outputs are specified. A frontier drawn for a single day with fixed machinery addresses a different question from a ten-year frontier that permits investment. An electricity generator's output set differs from the feasible electricity service of an entire interconnected grid. An institutional boundary is part of the definition of the object.
The standard introductory exposition in OpenStax (2022), production possibilities and social choices, illustrates the tradeoff between health and education. That representation is useful because the slope states forgone alternative output in physical units. It also makes the limitations of the elementary diagram visible: heterogeneous services, differing quality, complementary infrastructure, and uncertain outcomes must be incorporated if the geometry is to represent real social choices.
This investigation treats a frontier as a relation among possible outputs, not as a numerical forecast of aggregate output. Numerical frontiers can be estimated, but possibility is a counterfactual property. Observations reveal selected feasible activities; the unobserved alternatives require an engineering, mathematical, or statistical construction. A point drawn beyond an estimated boundary may therefore reveal a superior technology, a different production environment, an accounting inconsistency, or noise.
Boundaries, units, and time
Suppose a region produces grain and electric energy. Grain can be measured in metric tons, electricity in megawatt-hours, available freshwater in cubic meters, labor in hours, and installed capital in machine-specific service capacities. A frontier in annual quantities requires annual resource balances. The same region's hourly electricity reliability requires a finer temporal representation; one annual total does not establish system adequacy at each hour.
Outputs must refer to comparable goods. A completed medical procedure, a health outcome, and a quality-adjusted year of life represent different coordinates. Counting procedures as output may reward interventions without registering treatment quality. Grain tonnage without protein content can obscure differences in nutrition. The specification must reflect the scientific question.
The boundaries of possible choice also depend on the treatment of imports and exports, inventories, intermediate inputs, and waste. An analysis that adds steel produced to cars made from the same steel may count both an intermediate and a final product. Net output accounting resolves this when the question concerns final consumption. Gross output remains meaningful for a study of an industry's production process when the corresponding intermediate inputs are included.
A fully specified frontier establishes which comparisons are valid. The definition of attainable output is thus prior to optimization, efficiency assessment, and price interpretation. Altering this definition can alter the conclusion even when all recorded production quantities remain unchanged.
Historical foundations and analytical traditions
Costs, value, and production
Price theory connects the use of resources to alternative economic returns. In the historical Marshallian treatment, Principles of Economics (Marshall, 1920, Book V) situates supply and demand within processes of price adjustment. The costs relevant to a firm depend on the period and on which resources can be varied. This temporal insight carries into the distinction between fixed-capacity and long-run production frontiers.
Consider an artisan workshop with a limited number of machine hours. If it can sell two products at given prices, there is an immediate opportunity cost of allocating the next machine hour. If it can expand capacity over several years, the cost of that allocation changes. The source of sacrifice is the scarcity of resources over the relevant decision period, rather than a timeless technical coefficient.
Classical comparative advantage is expressed especially clearly through opportunity cost. The necessary comparison concerns the relative quantities forgone within one production unit and another. Comparative advantage can therefore coexist with absolute productivity superiority. This relation does not require a claim that labor is the only input. The general transformation-rate interpretation permits labor, capital services, land, energy, and intermediate inputs to constrain production simultaneously.
General equilibrium and the formal production set
Arrow and Debreu (1954), in Existence of an Equilibrium for a Competitive Economy, integrate production and exchange in a conditional competitive equilibrium existence argument. Their framework makes production possibilities and the assumptions governing producer and consumer behavior explicit. The historical contribution is essential to the interpretation of the frontier: the existence of a technologically possible bundle, the existence of an equilibrium, and the welfare properties of an equilibrium are separate propositions.
A closed feasible set with well-defined resource constraints allows its maximum to be characterized where compactness or suitable boundedness conditions hold. Convexity permits important separating and supporting hyperplane results. Neither existence of a frontier nor a physical interpretation of opportunity cost depends on all competitive equilibrium assumptions. The more demanding problem is whether a system of linear prices can support every efficient configuration under decentralized decisions.
In applications, the conditions of an abstract theorem are hypotheses. An industrial plant with fixed startup costs, an electric network with limited interconnection, or a patent-protected innovation may require a different market structure than the idealized competitive system. A theorem's logical validity does not transform its assumptions into descriptions of actual institutions.
From growth accounting to efficiency analysis
Solow (1957) explicitly separates movements along an aggregate production function from shifts of that function in technical change and aggregate production. His accounting method depends on assumptions about the relationship between factor remuneration and marginal products. The historical lesson is methodological: observed production growth must be decomposed using assumptions, not assigned to technology by inspection.
Farrell's (1957) measurement of productive efficiency established an influential line of production-efficiency research. The distinction between technical efficiency and cost-conditioned allocative efficiency is clarified in ten Raa's (2008) analysis of resource utilization and TFP. The distinction makes it possible to study how existing resources are utilized separately from how feasible production changes over time.
The resulting traditions address related but distinct objects. Price theory asks about choice among alternatives; general equilibrium asks about the consistency and support of many choices; measurement of technical efficiency compares observations with an estimated reference; growth accounting assigns measured changes to input growth and a residual; innovation economics explains why the technological opportunity set changes. A coherent account of frontiers brings these questions into one inquiry while preserving their distinct assumptions.
Production sets and the geometry of transformation
Formal representation
Let x be a nonnegative vector of m available inputs. Let y be a nonnegative vector of n desirable outputs. For a specified period and environment z, define the technology T(z) as the set of pairs (x,y) that can be jointly realized. The output possibility set is
The notation x can represent an upper bound on available inputs rather than the quantity actually consumed, provided that the technology permits unused inputs. Environmental conditions z can include soil quality, climate, grid connections, the set of permitted techniques, and other conditions held fixed. When these variables are varied, the analysis compares conditional sets.
Assume P is nonempty and compact for fixed x and z. An output y is strongly efficient if there is no feasible y′ satisfying y′ at least y in every coordinate and strictly greater in at least one. It is weakly efficient if no feasible y′ is strictly greater in every coordinate. Strong efficiency is a stronger requirement. On a horizontal portion of the upper boundary, some points may be weakly efficient but dominated. Hence the geometric upper boundary and the strongly efficient frontier may differ.
Under free disposal of desirable outputs, any nonnegative componentwise reduction of a feasible output remains feasible. This is an assumption about the possibility of withholding or discarding outputs. It is appropriate for many goods only when disposal costs and associated harms are accounted for. Waste that requires treatment is a productive obligation; its cost should remain inside the feasible set.
For two desirable outputs q and r, define the conditional maximum
The maximum exists for q in the projected domain if the feasible set is compact. Where the maximum is single-valued, the upper boundary can be drawn as r=g(q;x,z). If the set is downward closed, g is nonincreasing. With convexity, the feasible mixture of two outputs is available, and g is concave on its domain. The proof is direct: a convex combination of two maximal feasible points is feasible, so the maximum at the intermediate q is at least their interpolated r.
The familiar outward-bending frontier is therefore a consequence of particular technological and aggregation assumptions. It can be linear or piecewise linear; indivisibility can generate a nonconcave envelope; joint production can generate upper boundaries with different interpretations. The sign and curvature are results to derive from the feasible set.
Derivation from resource allocation
Take a single divisible resource L, split as l and L−l between two sectors. Let q=f(l), r=h(L−l). Suppose the sectoral functions are differentiable, strictly increasing, and concave. Along the boundary,
The marginal rate of transformation of q into r, expressed as units of r forgone per additional unit of q, is MRT = −dr/dq. As l rises, the marginal product in the q sector weakly declines, while the marginal product in the r sector weakly increases because its resource allocation declines. Their ratio consequently weakly rises where both derivatives are positive. This establishes increasing opportunity cost in this example without treating it as a universal fact.
For a hypothetical economy endowed with 100 input units, choose q=10√l and r=10√(100−l). The boundary satisfies q²+r²=10,000 in the nonnegative quadrant. Its slope is −q/r where r>0; the second derivative is −10,000/r³. At q=60, r=80 and MRT=0.75. At q=80, r=60 and MRT=4/3. The discrete opportunity cost between these allocations is (80−60)/(80−60)=1. The marginal and discrete concepts are distinct.
| Allocation l | Output q | Output r | MRT |
|---|---|---|---|
| 0 | 0 | 100 | 0 (one-sided) |
| 36 | 60 | 80 | 0.75 |
| 64 | 80 | 60 | 1.333… |
| 100 | 100 | 0 | Undefined as a finite slope |
All quantities in this example are hypothetical. They illustrate the resource-allocation mechanism and cannot be interpreted as empirical estimates of real inputs or output substitution.
Multiple constraints and shadow values
Let a vector a denote activity levels. Let A contain required resource quantities per unit of activity and B contain activity output coefficients. Consider the feasible conditions a≥0, Aa≤x, and y≤Ba. When the activities are continuously divisible, the attainable set is convex. With all coefficients specified, maximizing a weighted output vector is a linear program.
At a solution, dual values for the resource inequalities express the marginal value of additional input under the selected objective, when that marginal value is well defined. A constraint may have a zero shadow value even though it appears important physically: other constraints may prevent productive use of additional supply. Conversely, a resource with a small physical quantity may be the bottleneck.
For example, suppose one activity requires two labor units and one power unit per unit of output, while another requires one labor unit and two power units. If labor and power are each limited to 12, feasible output quantities q and r satisfy 2q+r≤12 and q+2r≤12. The intersection is (4,4); the intercepts are (6,0) and (0,6). The upper boundary has a kink because the binding resource changes. The tradeoff is −1/2 on the segment from (0,6) to (4,4), and −2 from (4,4) to (6,0). A single differentiable curve would conceal the switch in resource scarcity. The example can be solved directly by substitution.
The shadow value of labor depends on output weights. For an objective q+r, both constraints bind at (4,4) and the dual values are 1/3 for each resource. Increasing labor alone by one unit, leaving power fixed and retaining the same binding structure, changes q by 2/3 and r by −1/3. The objective rises by 1/3. The resulting shadow value is conditional on the specified weights and local structure, not an unconditional market wage.
Opportunity cost and decentralized prices
From physical tradeoffs to economic cost
A physical MRT is a ratio of quantities. To express it in monetary terms, multiply forgone output by its relevant price. That price must correspond to the output definition and the relevant time and market. A lost megawatt-hour in a congestion-constrained network at peak demand may have a different economic value from an off-peak megawatt-hour.
Let p and s be positive prices of q and r. The producer with fixed available resources solves max pq+sr over P. If r=g(q) and an interior solution is differentiable, the first-order relation is
Concavity of g makes this condition sufficient for a global maximum on the feasible interval. At nondifferentiable points, a supporting price ratio may fall within the range of one-sided slopes. At endpoints, the relevant conditions are inequalities. Positive prices ensure that a maximizer cannot be strongly dominated by another feasible output combination.
For the circular frontier q²+r²≤10,000 with p=s=1, the optimum is q=r=100/√2, yielding revenue 100√2. If p=2 and s=1, the maximum occurs at q=200/√5, r=100/√5. The MRT is 2. This allocation shifts toward q despite no change in the physical technology. Observed changes in output mix following price changes are therefore insufficient to identify a shifted production set.
Prices, welfare, and distribution
An interior social optimum for a differentiable increasing objective W(q,r) on the frontier has Wq/Wr=MRT. The equality identifies a tangency conditional on the welfare objective. Market prices can play a supporting role when the institutional and competitive conditions permit them to reflect the relevant marginal valuations. Private revenue maximization and social welfare evaluation are distinct exercises.
For health and education, even a carefully specified physical transformation frontier cannot establish which allocation best serves a society. Objectives may involve equity, minimum service entitlements, geographic access, and risk. With such constraints the selected point may be at a corner or within a two-dimensional projection while efficient in a richer representation. Pareto efficiency in production is a statement about output dominance, not an ethical ranking of complete social arrangements.
At the international scale, the price ratio that supports a national production plan can reflect world demand, domestic market power, trade costs, and policy. Transfers alter feasible consumption or income while leaving the short-run production technology unchanged. The choice to compensate those losing from reallocation is an independent distributive and institutional decision. Aggregate surplus does not itself establish that compensation occurs.
Finite changes and changing objectives
Marginal transformations are local. A multi-year capital project consumes resources across periods and changes the relevant opportunity set. The sacrifice from moving between distant allocations is measured as the total change in alternative output, not as the initial marginal rate multiplied by the whole change unless the rate is constant. On a smooth boundary this discrete sacrifice equals the integral of the changing MRT over the interval.
With q increasing from 60 to 80 in the preceding circular example, the lost r is 20. Applying the initial MRT of 0.75 to all 20 units would give 15, understating the sacrifice. Applying the final MRT gives about 26.67, overstating it. This elementary calculation explains why project assessment needs the full transformation schedule when marginal rates vary materially.
Nonconvex technologies and joint production
Indivisibility and unsupported efficient choices
Production configurations may be discrete because machines, software systems, storage plants, or entry facilities require fixed startup commitments. Suppose an indivisible facility can choose at most one of three output configurations per period: A=(0,10), B=(4,4), and C=(10,0). Include all componentwise smaller nonnegative bundles for each configuration. All three displayed points are strongly efficient. B cannot be supported by any nonzero nonnegative price vector.
Indeed, supporting B requires 4p+4s≥10s and 4p+4s≥10p. The first implies p≥1.5s, the second s≥1.5p. Both hold only at p=s=0. Thus an efficient point may have no supporting nonzero linear price even with free disposal. A convexification of the set would include (5,5), which dominates B. That point is unattainable under the stated indivisibility.
If identical facilities can be replicated at sufficient scale or operated in separate periods with costless switching, convex mixtures may become feasible in an aggregate or time-averaged sense. The economic distinction hinges on whether replication and time sharing are genuinely available within the definition. A convex hull is not an empirical observation; it is a mathematical expansion that needs a technological interpretation.
Nonconvexity also matters for cost recovery. In an illustrative single-output technology with cost C(q)=F+cq for q>0 and C(0)=0, marginal cost is c, while average cost is c+F/q. Pricing each positive unit at marginal cost fails to recover the startup cost from sales. The institutional implications may include fixed tariffs, contract arrangements, or other financing. Which arrangement is efficient or equitable depends on demand, strategic behavior, and constraints.
Joint production and undesirable outputs
Many production activities generate co-products. Refineries refine multiple fuels; farming produces food, residues, and emissions; hospitals generate treatments together with resource-consuming waste. A technology allowing the simultaneous production of q and r can exhibit complementarity through common inputs even if its downward-closed maximum envelope is nonincreasing.
Suppose a plant produces q units of useful output and b units of waste, with b=q under its existing technology. A second technique produces the same q with b=q/2 but needs additional equipment. If emissions or waste are undesirable outputs, productive improvement means raising desirable outputs or reducing undesirable outputs under feasible constraints. Treating b as an ordinary freely disposable good would incorrectly allow the plant to set reported waste to zero without treatment.
The feasible relation can represent weak disposal by allowing a joint proportional contraction of desirable and undesirable outputs. Under the specific assumption that each feasible (q,b) can be replaced by (θq,θb) for θ between zero and one, reducing waste by scaling down activity also reduces useful output. Additional abatement technologies may expand the relation. Weak disposability is a model assumption, not a universal law of pollution: some pollutants can be separated with distinct resources, while others require process redesign.
The efficient boundary for mixed desirable and undesirable outputs is defined using the appropriate order: improvement increases desirable coordinates and decreases harmful ones. When resource consumption, quality, and pollution are treated separately, an apparently interior two-output projection can correspond to an efficient higher-dimensional allocation.
Risk, fixed commitments, and feasibility
A production plan may be feasible in normal operation and infeasible under a disruption. In an electricity system, weather-dependent generation, fuel supply, maintenance, and reserve requirements make feasibility state dependent. One can represent a set Pω for each state ω, or a robust set consisting of plans feasible across specified states, provided that plans and adaptable decisions are distinguished.
A factory holding reserve equipment may achieve smaller average current output than the highest observed performer. The reserve has option value if downtime is costly. A frontier estimated using only current output may classify resilience as inefficiency. The model should record the reliability requirement or the state-contingent outcomes rather than infer waste from the projection.
Technological change and directional tradeoffs
Set inclusion and heterogeneous improvement
A technological expansion occurs if P(x;z₀) is contained in P(x;z₁) for fixed input availability and comparably defined outputs. Strict inclusion means at least one newly attainable combination. A uniform radial expansion, P₁=aP₀ for a>1, is a particular case. Innovation may instead be sector-specific, input saving, quality enhancing, or pollution reducing.
Suppose the earlier frontier is q²+r²≤10,000 and a new technique doubles the attainable q at each r without changing the r technology. The new boundary is (q/2)²+r²≤10,000. The q intercept expands from 100 to 200; the r intercept stays at 100. At q=60 the old r maximum was 80, while the new r maximum is √9,100, approximately 95.39. The new MRT at q=60 is q/(4r), approximately 0.157. The same q level therefore requires a smaller marginal sacrifice of r.
At the same proportional allocation, q=120 and r=80 replace q=60 and r=80. The transformation rate is 0.375 rather than 0.75. These calculations isolate the specific geometric effect of sector-biased improvement. They do not imply that every relative price or factor income must move in one direction because equilibrium demand and the distribution of resource ownership also matter.
A new technology may not create nested short-run sets if adoption requires dismantling old equipment, temporarily withdrawing labor for training, or losing access to an input. The portfolio of technically known processes can expand while the locally operational choice set contracts during conversion. The relevant distinction is between knowledge, installed capability, and available operating resources. Each is a state variable defining a different attainable set.
Factor bias and the direction of research
A production function Q=Af(K,L) is Hicks-neutral with respect to input substitution at fixed input ratios: A changes the scale of output while the ratio of marginal products is unaffected. A factor-augmenting specification Q=f(AK K, AL L) changes effective inputs. Its effect on the marginal rate of technical substitution and factor shares depends on functional form, substitution, and the point of evaluation.
The direction of innovation responds to incentives. Acemoglu (2002), in Directed Technical Change, analyzes competing price and market-size effects. His formal setting explains why the relative scarcity or prevalence of a factor can influence research incentives in different directions. The result is conditional on the model and the elasticity of substitution; it cannot be reduced to a universal rule that innovation always economizes on the most expensive input.
A sector-level output bias and an input-level factor bias are related but nonidentical. A machine can reduce skilled labor per unit of a particular product, thereby expanding the product's feasible output, while increasing demand for skilled maintenance work elsewhere. Identifying a skill-saving process requires measuring the complete relevant production system and factor use, rather than observing only employment after adoption.
Knowledge, replication, and market structure
Romer (1990), Endogenous Technological Change, treats technological improvement as an outcome of investment and explains the importance of ideas that can be used by multiple producers under partial exclusion. The nonrival character of knowledge changes the relation between the cost of creating a design and the cost of replicating it. Consequently, the social benefit of developing a process can differ from the return an individual innovator can appropriate.
A digital design may be transmitted at low incremental cost, yet its implementation can require costly capital, compute, energy, data, and organizational integration. Knowledge availability is thus insufficient to establish productive feasibility at a specific site. The distinction is particularly important for comparisons across economies with unequal financing, infrastructure, skills, and access to protected technology.
Let an invention increase the maximum attainable q in a mature facility while requiring an initial investment. An operating frontier conditional on installed equipment records the higher output capacity. An economy-wide frontier including the resources that produced the equipment also records the opportunity cost of adoption. These descriptions are consistent because they condition on different resource stocks and time horizons.
Intertemporal production and adjustment
Investment as a use of present resources
A static frontier places current outputs under current resources. A dynamic production problem includes outputs that augment the resources available in later periods. Let Cₜ be consumption and Iₜ investment. In a hypothetical one-good system,
The feasibility of future output depends on the current investment decision, depreciation δ, and future technology and labor. With nonnegative consumption and investment, increasing Iₜ has an immediate physical opportunity cost of current Cₜ at a fixed current production level. The future marginal benefit of investment depends on the next period's marginal productivity of capital.
For example, set current output to 100, capital to 100, depreciation to 0.1, labor fixed, and next-period output equal to twice the square root of next-period capital. Investment 10 leaves consumption 90, restores capital to 100, and yields future output 20. Investment 30 leaves consumption 70, raises capital to 120, and yields future output about 21.91. The second plan exchanges 20 units of current consumption for about 1.91 additional next-period output, with remaining capital and any later benefits outside this deliberately truncated example. Evaluation requires a horizon, a terminal capital treatment, and a valuation rule. The numbers are illustrative and are not calibrated.
Capital and technology are distinct determinants of dynamic feasibility. A higher capital stock can expand output under unchanged knowledge. Innovation can also augment the productivity of existing capital or introduce new goods. An empirical comparison that holds nominal investment constant may still involve changing real capital services, asset quality, and depreciation.
Adjustment costs and installed capabilities
A firm that replaces an operating line often incurs temporary setup time, learning losses, and stranded assets. If output in the conversion period falls, the decline can occur alongside a long-run frontier expansion. Intertemporal feasibility is represented by entire sequences of outputs and resource commitments. A static comparison across two dates cannot capture the cost of intermediate transitions.
Consider a three-period hypothetical choice. Operating the old technology yields 10 each period. Switching now yields 3, then 14, then 14. The old plan yields 30 units in total and the new plan yields 31. For equally valued units and a discount factor β, the difference in present quantity equivalents is −7+4β+4β². At β=1 the difference is 1. At β=0.9 it is −0.16. This example isolates the role of timing in the assessment of a technically superior steady state. In a longer horizon the ranking can change again.
Discounting in a social analysis can involve contested valuations across generations. The production relation identifies feasible paths; the discount factor selects a ranking over those paths. This separation is important when public infrastructure, environmental mitigation, or research generates benefits extending beyond private planning horizons.
Irreversibility and uncertainty
A technology adoption decision can be irreversible within the relevant period. If future demand or input costs are uncertain, waiting may preserve an opportunity to select a better process after information arrives. A feasible policy is then a contingent decision rule, not merely a deterministic output vector. The value of a flexible policy must be compared with the cost of maintaining optionality.
An observed firm operating below instantaneous full capacity may hold spare capacity for uncertain demand. Whether the firm is economically inefficient depends on the information available at decision time and the objective. Comparing its expected output with a reference that assumes perfect foresight biases the analysis against feasible risk management.
A multi-period feasible set can include maintenance, learning, research, and investment as activities. Some produce measurable output only after delay. A research program whose discoveries are uncertain is naturally represented as an activity with state-dependent returns and resource costs. Innovation expands future opportunities probabilistically; observed patent counts or spending on research alone do not establish the magnitude of the expansion.
Efficiency, productivity, and frontier estimation
Technical efficiency and distance
The distance of an observed output from a reference set measures a conditional shortfall. For nonzero y in P(x), define the output expansion factor
Under boundedness and feasibility, φ≥1. Output-oriented radial technical efficiency is 1/φ. A value of 0.8 means the observed vector can be proportionally expanded by a factor of 1.25 at the same specified inputs if the reference frontier is accessible. Input orientation instead asks what proportional contraction of inputs could maintain the specified outputs. Under nonconstant returns to scale, the two orientations generally differ.
Radial efficiency does not resolve every form of slack. Let the feasible region be the rectangle 0≤q≤10 and 0≤r≤10. The observed point (10,5) has φ=1 because q cannot expand. Yet r can increase to 10 without sacrificing q. The point is weakly but not strongly efficient under the defined order. A radial efficiency score must therefore be accompanied by slack analysis where the scientific question concerns componentwise dominance.
Physical efficiency differs from minimum cost conditional on input prices and output targets. Suppose y can be produced either using inputs (1,4) or (4,1), with no feasible input reduction from either technique. If input prices are (2,1), their costs are 6 and 9 respectively. The second technique is technically efficient by input dominance but costs more. An allocative judgment depends on the price vector; a welfare judgment further depends on the institution and valuation of inputs.
Deterministic peer-based frontiers
Charnes, Cooper, and Rhodes (1978) developed multi-input, multi-output efficiency measurement based on observed decision-making units and optimization. A standard constant-returns-to-scale input-oriented envelopment problem for unit o is
Here X and Y contain the observed input and output vectors of the reference units; λ selects a nonnegative combination. A variable-returns specification adds the constraint that the λ weights sum to one. Constant returns extends combinations along rays, whereas variable returns limits them to convex combinations. The analyst must justify the assumptions for the production unit, and treatment of slack remains important.
A simple hypothetical peer exercise illustrates the distinction. All units use one input and produce one output. A uses 10 and produces 10, B uses 20 and produces 30, C uses 15 and produces 15. With constant returns, B's ratio 1.5 is highest, so C could produce 15 with 10 input units; its input-oriented score is 10/15=2/3. With variable returns, the line segment from A to B reaches output 15 at input 12.5, so C's score is 12.5/15=5/6. This assumes free disposal and the stated three observations; both results are relative to the constructed reference set.
Neither score establishes why C differs. Its product quality may be better, its operating environment may be more difficult, or its managers may face credit constraints. Peer-based measurements are particularly sensitive to outliers and to the number of output and input dimensions relative to the sample. A frontier formed from a small sample of heterogeneous entities should not be treated as the universal technical limit.
Stochastic frontiers
Aigner, Lovell, and Schmidt (1977), in stochastic frontier production models, separate an idiosyncratic noise term from a one-sided inefficiency component. A common log-output representation is
If f is a positive frontier output and u enters in this form, exp(−u) is technical efficiency relative to the conditional deterministic component. A positive shock v can place observed output above that component. The statistical interpretation relies on assumptions regarding the distributions and dependence of v and u, specification of f, and measurement of x and y. Estimated inefficiency is latent; it is not simply the observed residual.
Data selection, omitted inputs, unobserved environment, and technological heterogeneity can all influence residual asymmetry. A stochastic frontier may interpret such heterogeneity as inefficiency when the model fails to represent it. Panel designs can help distinguish persistent differences from transient shocks, but they do not solve identification automatically. A structural assumption about variation remains necessary.
Deterministic and stochastic methods offer complementary evidence. An engineering benchmark isolates known physical capacity; a peer envelope reveals comparable observed practices; a stochastic model adds a distributional account of measurement variation. Disagreement can indicate a mismatch in technology, quality, constraints, or functional form. It is often more informative to examine that disagreement than to report a single ranking.
Growth accounting and the frontier
Under a differentiable aggregate production representation Q=A f(K,L), proportional changes satisfy
The output elasticities equal the corresponding competitive income shares under marginal-product remuneration with appropriate accounting. Standard growth accounting also uses assumptions about returns to scale and output and input aggregation. The residual can then be measured by subtracting weighted input growth from output growth. It is not automatically a pure measure of inventions or a direct mapping of the multi-output frontier.
The OECD productivity manual (OECD, 2001) establishes the measurement context across aggregates and industries. The U.S. Bureau of Labor Statistics productivity concepts handbook makes clear that measured multifactor productivity can reflect changes in organization, utilization, resource allocation, scale, and technology that are not independently recorded on the input side. The same statistical formula can aggregate heterogeneous mechanisms.
For a hypothetical period, let measured log output growth be 0.04, capital services log growth 0.06, labor input log growth 0.01, and shares 0.4 and 0.6. The measured residual is 0.04−0.4×0.06−0.6×0.01=0.010, or one log percentage point. A revision that raises measured labor input growth to 0.02 reduces it to 0.004. Nothing in the data example directly observes an invention.
Capital services are analytically important because the productive contribution of assets may differ from the value of the asset stock. Labor hours differ from labor input adjusted for composition. Revenue output changes when prices or quality change. A frontier comparison must distinguish physical quantity improvement, changes in input quality, and changes in the value of product mix.
Separating catch-up from change
For a single output under fixed resources, suppose an initial feasible maximum is 100 and actual output is 80. A later maximum is 120 and actual output is 108. Initial efficiency is 0.8, later efficiency 0.9. Actual output rises by 35%; the frontier rises by 20%; the efficiency ratio rises by 12.5%. Multiplicatively, 1.2×1.125=1.35. Additively in logs, the corresponding changes sum exactly.
This arithmetic identity is valid given two comparable maximum outputs and unchanged inputs. It illustrates the kind of distinction pursued in frontier productivity research, including ten Raa (2008). Measuring a true maximum is a substantive challenge. A change in the reference sample, resource availability, or definition of output could create the same accounting pattern without technological progress.
For multiple outputs, no unique scalar rate of frontier change describes all possible directions. Distance functions and productivity indexes select a direction or reference weights. A Malmquist-type comparison can evaluate observations using reference-period technologies, with appropriate assumptions about orientation and feasibility of cross-period distances. Different index constructions may give different numerical decompositions. The interpretation should disclose those choices, particularly when frontiers cross or the product composition changes.
Open economies, networks, and aggregation
Comparative advantage and consumption feasibility
Assume two hypothetical economies, A and B, with equal labor endowments normalized to one unit. A can produce up to 12 units of food or 6 units of machinery; B up to 4 food or 4 machinery. With constant unit labor requirements and divisible production, both frontiers are linear. A's cost of one machinery unit is two food units; B's cost is one food unit. A has absolute advantage in both products, but B has comparative advantage in machinery.
At an illustrative exchange rate of 1.5 units of food per machinery unit, A specializes in food and B in machinery. If they exchange three food for two machinery, A consumes (9 food,2 machinery) and B consumes (3 food,2 machinery). Individual production frontiers obey q+2r≤12 for A and q+r≤4 for B. The consumption bundles have left-hand sides 13 and 5 respectively. Both are outside their autarky feasible production sets. World production is (12,4), and consumption sums to (12,4), so the example respects material balance.
The distinction between national production and consumption possibility sets is essential. Trade can allow consumption beyond a domestic frontier while world production remains within the aggregate feasible set. The gains in this stylized example depend on constant relative costs, absence of trade frictions, and the stipulated achievable exchange rate. In actual economies, terms of trade, adjustment costs, distribution, supply-chain security, and market power alter the feasible gains and their allocation.
International investment may also change the production frontier through capital formation and access to techniques. The distinction between a larger trading opportunity set and a technologically expanded production set remains meaningful. Both can change simultaneously, requiring a counterfactual that holds the other mechanism fixed.
Networks and bottlenecks
Intermediate inputs connect sectors. A downstream producer may have unused machines because an upstream input is unavailable. Consider a network with component output capacity 50, assembly capacity 80, and one component required per final product. Final output is at most 50. Increasing assembly capacity to 100 does not change this bound. Increasing component capacity to 70 raises final possible production to 70, provided other inputs are sufficient.
This simple example prevents the inference of aggregate technological progress from isolated capacity improvements. A local frontier can expand without shifting the final-output frontier. Conversely, reorganization of sourcing can expand feasible final deliveries without a new production technique at either factory. Network structure is a condition of joint feasibility.
Mathematically, aggregate possibilities can be formed from the sum of individual net output sets under resource balances and transfer conditions. If all producers have independently available resources, the Minkowski sum gives joint feasible outputs. If they share energy, water, labor, or intermediate inputs, simple summation without the joint constraints overstates feasibility. A common resource can be counted only once.
An input-output representation with intermediate requirements A and final demand d satisfies gross activity output q≥Aq+d, along with capacity and factor constraints. If production coefficients are nonnegative and the spectral radius of A is below one, the usual inverse (I−A)⁻¹ exists and is nonnegative. The implied activities required by a final output vector must still satisfy available capacity and resources. The algebraic existence of gross requirements does not establish that they can actually be supplied.
Aggregation and institutional capacity
A country-wide frontier is more than the sum of enterprise-level output maxima. It requires decisions about the mobility of labor and capital, transport costs, interregional infrastructure, and how alternative uses interact. Local unemployment may coexist with a shortage of specialized labor in the expanding sector. Aggregating worker-hours as if they were identical can imply a reallocation that requires training or relocation.
Similarly, financing can constrain attainable production for a firm even when physical resources exist elsewhere. The national physical frontier may include a configuration that is infeasible for a firm facing a binding borrowing limit. A conditional firm-level feasible set can include financing constraints, while a national engineering set may abstract from them. Researchers should identify which question the frontier answers before classifying the difference as technical inefficiency.
An observed revenue productivity difference can reflect differing markups, demand, quality, and input prices. Physical productivity differences concern quantities for defined inputs. A cross-national study using only sales in converted currency mixes these problems unless suitable deflators and comparisons are provided. Local price variation and purchasing-power adjustments answer different statistical questions.
Environmental and sectoral evidence
Agricultural attainable yields
Agricultural production provides a concrete domain in which frontier measurement is conditional on biophysical environments. Mueller et al. (2012), Closing yield gaps through nutrient and water management, study 17 major crops and compare observed production with attainable yields in comparable climate zones. Their approach links yield limitations to regional variation in management, water availability, and nutrients. The study is observational and model-based, not a randomized estimate of a universally attainable increase.
The authors subsequently published a 2013 corrigendum concerning the article. The correction record matters because geographic interpretation of estimated gaps depends on mapping and data processing. This dossier does not reproduce the affected regional map, infer corrected estimates, or treat older estimates as current regional statistics.
Yield per hectare is a partial productivity measure. A higher yield can depend on greater irrigation, nutrient use, energy use, labor, or capital. Under a water constraint, the relevant attainable set may describe crop output jointly with water withdrawal or water consumption. Under an ecological objective it may also include nutrient runoff and habitat loss. The frontier changes if the researcher switches from land productivity to a multi-input, multi-output comparison.
Suppose a hypothetical farm has 100 hectares and can produce 300 tons of grain with 1,000 water units. A second method produces 360 tons using 1,400 water units. Land productivity rises 20%, but grain per water unit falls from 0.30 to approximately 0.257. Whether the new method enlarges the attainable set under the original water budget depends on the availability of divisible activity, output-water response, and other inputs. The reported output growth alone does not establish joint resource efficiency.
A third method producing 330 tons with 900 water units and no additional unspecified inputs would dominate the first in these measured coordinates. Before estimating the economic significance of that dominance, one must compare yield stability, quality, labor, costs, and feasibility of adopting the method. In farming, land and water constraints are spatially specific; scale-up and aggregate environmental impact require explicit treatment.
Electricity production and energy system boundaries
Electricity is often compared using the levelized cost of electricity (LCOE), a discounted lifetime cost per discounted lifetime electricity output under stated assumptions. The International Renewable Energy Agency's (2025) Renewable power generation costs in 2024 identifies the relevance of capacity factors, market mix, and financing in its reported changes. This dossier uses that qualitative observation, not an updated cost series or a causal estimate from the report.
A stylized discounted cost calculation is
where C includes specified investment and operating costs, E is generation in the defined periods, and i is the discount rate. Different conventions exist for taxes, financing, decommissioning, and other costs. The measure is conditional on its boundaries. It is not a direct multi-output physical transformation frontier.
A low generation cost can coexist with the requirement for complementary flexible supply, storage, transmission, reserves, and capacity to meet demand when production is low. System feasibility relates electricity delivered at times and locations to network constraints and reliability requirements. A project-specific average cost cannot measure all those system services.
For illustration, a hypothetical isolated system requires 10 units of power in each of two periods. One production process supplies at most 20 units in the first period and none in the second. Without storage or backup it cannot meet demand in both periods despite producing enough energy in total. With lossless storage capacity of 10, matching power capacity, and otherwise sufficient generation, the system can produce 20 in the first period, deliver 10 immediately, and store 10 for the second. Storage expands the feasible service configuration, even though the generator's technology is unchanged.
Climate policy adds a constraint expressed as a cumulative emissions budget, or as period-specific emissions limits. The IPCC (2022) mitigation pathways assessment organizes scenarios around differing strategies and assumptions. Model pathways indicate conditional feasible trajectories under the model, not guaranteed outcomes for any one country. Their economic meaning includes technological choices, capital turnover, resource availability, and constraints on harmful outputs.
Manufacturing, services, and the empirical unit
A manufacturer can increase physical output per working hour through better machinery, improved scheduling, longer equipment utilization, or product mix. All alter observed productivity; only some require a change in the set of available technologies. A factory sample should measure operating hours, quality, machine services, energy, materials, interruptions, and the availability of intermediate inputs wherever relevant.
The World Bank Enterprise Surveys methodology documents a cross-country formal-sector survey with stratified sampling and a defined universe of eligible firms. Its formal sample and industry coverage should not be equated with all production, particularly in economies where informal enterprises account for a substantial but separately measured segment. Panel and weighting conventions matter for cross-country inference.
Service output creates an additional measurement problem. A school can produce completed instructional hours, measured learning, and access for disadvantaged students. A hospital can produce episodes of care and improvements in health. Counting volume without quality can create an apparent frontier expansion through lower service quality. Including these dimensions changes which units are comparable and how output dominance is defined.
In public services, resource constraints and minimum access requirements shape feasible alternatives. A remote clinic may use more resources per treatment than an urban clinic while providing access for people with otherwise limited alternatives. The frontier of service delivery conditional on population distribution differs from an envelope using treatment counts alone. The suitability of a benchmark thus depends on whether the decision unit faces comparable obligations.
Empirical identification and comparative research
Observable implications of alternative mechanisms
Several distinct events can generate the same observation: higher output between two dates. A factory may acquire more machines, change the use of existing machines, shift toward a high-value product, adopt a better process, remove a financing constraint, or reduce measured quality. A research design should formulate discriminating predictions before interpreting the outcome.
| Mechanism | Required comparison | Potentially discriminating observation |
|---|---|---|
| Resource accumulation | Same technique, different available inputs | Measured capital services or labor input expand |
| Reallocation | Same conditional feasible set | Output mix changes with relative prices or resource use |
| Technical catch-up | Comparable reference set | Distance to benchmark narrows at measured inputs |
| Frontier expansion | Comparable inputs, outputs, and environments | New attainable combinations beyond previous benchmark |
| Constraint relaxation | Fixed underlying technique, different constraints | Financing, materials, water, or network bottleneck demonstrably relaxes |
| Measurement change | Same activity, revised statistical definitions | Growth changes when consistent series are reconstructed |
The table is an analytical guide, not an identification theorem. The mechanisms can occur jointly. A measured increase in capital services may embed new technology; innovation can be accompanied by improved managerial capability; a financing reform can enable adoption of a new process. Empirical disentanglement requires a research design capable of observing or instrumenting those mechanisms.
A reproducible design for cross-country comparisons
First define the unit: establishment, firm, industry-region pair, or national economy. Choose a common reporting period and document differences in price levels, output composition, product quality, accounting conventions, and productive environments. Quantity measures are preferable for physical tradeoff estimation when outputs are homogeneous. Harmonized value-added and gross-output measures serve different research purposes.
Next identify resources and conditions. For an industry study, collect labor hours and composition, capital services, intermediate inputs and energy, output quality, capacity utilization, and exposure to shocks. Distinguish fixed environmental variables from resources that can be varied by the decision unit. Assign uncertainty intervals to critical quantities. Assess whether data are representative of the population under the relevant survey methodology.
Specify a frontier model before ranking units. A convex envelopment benchmark assumes access to combinations of observed practices. Where plants cannot be replicated or mixed, add a nonconvex specification as sensitivity analysis. A stochastic specification requires a credible functional form and noise treatment. Where data permit, compare engineering or process-specific benchmarks with statistical estimates.
For a longitudinal study, freeze reference-period definitions, calculate distances using clearly stated technologies, and examine the effect of entry and exit. Re-estimate after excluding outliers or treating environmental differences separately. Report whether conclusions survive alternative scale assumptions, output weights, and orientations. A shift based only on one unusually productive observation warrants special scrutiny.
Validate external comparability before forming a common global frontier. When countries use distinct technique families, a pooled frontier can estimate distance to the best performer conditional on the specified dimensions, but it need not describe each country's locally accessible options. Common and group-specific references can answer complementary questions: the first examines a shared benchmark, the second local performance conditional on accessible technology. The distance between references needs independent interpretation.
Finally, connect results to institutions. Document infrastructure, financing, regulatory requirements, trade access, and adjustment costs. A policy study should distinguish feasibility under existing conditions from feasibility after the proposed policy. Improvements in measured efficiency associated with a policy can arise through sample selection, demand changes, or shifts in prices. A causal claim requires more than a frontier score.
Identifying the effect of innovation
A credible adoption study needs a counterfactual: what outputs and inputs would have been observed had comparable units not adopted the innovation? Before-and-after comparisons are insufficient when adoption responds to expected demand or managerial ability. Randomization, credible natural experiments, or a justified quasi-experimental design may help, depending on the setting.
Difference-in-differences analysis requires assumptions about untreated outcome trajectories. Anticipatory investment, spillovers, heterogeneous timing, and changes in product quality can undermine a naive specification. An instrumental-variable design requires a source of variation affecting adoption and a defensible exclusion condition for the outcome. Engineering performance tests can establish technical improvements under controlled conditions while leaving economy-wide adoption costs and market responses unresolved.
Measure the effect of technology on multiple feasible configurations when possible. An intervention may reduce inputs needed for a standard product without expanding demand or measured output. It may enable a new quality tier that cannot be compared using legacy units. A design that observes only revenue growth may miss genuine physical improvements or overstate them because of price changes.
In a frontier estimation context, a causal effect on actual production does not automatically identify an effect on the frontier. The treatment could improve utilization of existing capacity. A study claiming innovation-induced frontier change should document a new attainable combination or reduction in required inputs at comparable output and quality, along with evidence against the relevant alternative explanations.
Validation and uncertainty
Frontier estimates are often more sensitive than estimates of average relationships because their object is an extremal surface. Data censoring, inaccurate high values, changing product definitions, or unrecorded subcontracting can move an empirical boundary. Uncertainty should accompany scores and changes where estimation permits it. A point estimate with an unreported measurement convention invites false precision.
For a multi-output study, report both the direction of comparison and any weight restrictions. Some outputs are not reducible to market prices, particularly public services and environmental outcomes. A weight chosen through an optimization procedure may have little policy meaning if it gives negligible value to an essential output. Minimum-output or quality constraints can capture obligations more transparently than a single unrestricted scalar objective.
The empirical materials reviewed here illustrate methods and measurement challenges, rather than furnish a newly harmonized cross-country dataset. The numerical examples are explicitly constructed to verify logical relationships. They do not establish magnitudes for actual economies. The empirical work appropriate to a specific technology requires data on the relevant production environment, causal research design, and sensitivity analysis.
Analytical synthesis and limits
What a frontier comparison can establish
The principal analytical conclusion is that the meaning of a production frontier is inseparable from its conditioning variables. A change in attainable production can be demonstrated only for a comparable resource bundle, set of outputs, technological horizon, and institutional boundary. Once that condition is fixed, the comparison between old and new sets can establish inclusion, contraction, crossing, or changes in feasible tradeoffs.
The shape of a frontier reveals scarcity relationships generated by a production specification. The slope of a differentiable two-output boundary gives a local physical transformation rate; at a kink, a range of supporting ratios may apply. The support of an efficient point by market prices depends on additional geometric and institutional conditions. Nonconvex points can resist linear price support even though they are physically efficient.
The examples demonstrate four distinct conclusions. The circular boundary establishes increasing sacrifice under concave allocation technologies. The two-resource linear system locates a kink caused by a change of bottleneck. The indivisible facility establishes the failure of full price support under nonconvexity. The directional expansion demonstrates that improving one sector can flatten the transformation schedule at fixed output levels without a proportional increase in every output.
These conclusions are derivations from standard economic concepts and defined hypothetical constructions, not claims to discovery of unknown mathematical theorems. Their value is the joint interpretation of physical possibilities, observed utilization, prices, technological investment, and institutional availability.
Conditional and aggregate feasibility
The most important practical distinction concerns the gap between an outer benchmark and an implementable plan. A technology observed in one economy may require capital, specialized skills, input networks, or contractual access unavailable elsewhere. Statistical distances can represent useful measures of disparity while leaving the cost of bridging the gap unmeasured. The same observation may correspond to local constraint, technological disparity, or managerial inefficiency depending on the benchmark.
For example, a water-limited region can compare farms with a physically similar climate and separately compare their access to irrigation and fertilizer. A manufacturing cluster can examine plant productivity and the reliability of its component network. An electricity system can compare generation cost and the feasible delivery of power across times. In each case the scientific object improves when the complementary conditions become explicit.
The interpretation of policy is consequently conditional on what a policy changes. A policy that reallocates resources selects a point within a feasible set. A policy that improves coordination can make previously known practices implementable. A public research program can create new knowledge and expand the technological set. An environmental limit can contract the permitted set in selected directions while inducing innovation that changes later feasible alternatives. One policy may operate through several of these channels.
Welfare and institutional choice
Neither a maximal product bundle nor a low measured input requirement supplies a complete welfare criterion. Production choices distribute goods, risks, income, and environmental consequences across households, locations, and generations. Some effects form additional output coordinates; others enter the social evaluation of outcomes and resource claims. A production frontier can clarify scarcity and feasible improvements without selecting the social objective.
The case of an efficient point unsupported by linear prices illustrates why institutions matter. Market allocations may depend on contracts covering fixed costs, public infrastructure, organization of pooled resources, or regulations governing harmful outputs. Different institutions can affect both the production set actually accessible and the allocation selected. Economic science evaluates these mechanisms by their assumptions, incentives, observable implications, and welfare conditions.
The relevant inquiry thus begins with physical and institutional feasibility, proceeds to comparative transformation and production costs, and only then examines selection among alternatives. This ordering protects a technological conclusion from being confused with a price movement or welfare judgment.
Limits and unresolved scientific questions
No single study can estimate the complete global production possibility frontier. Output dimensions are numerous, quality varies, and technologies evolve. A theoretical representation can identify conditions and consequences, but quantitative conclusions depend on a defined empirical unit and measurement conventions. The present monograph provides formal explanatory examples and a comparative research design; it does not claim a measured global technology surface, cross-country efficiency ranking, or novel identification theorem.
The main unresolved empirical questions concern comparability of service quality, valuation and physical measurement of intangible outputs, identification of jointly produced harmful outputs, and measurement of feasible options under incomplete access to technology. Further research can compare alternative production references, estimate adoption costs and transition paths, and investigate how market and nonmarket institutions influence the observation of technical progress.
A particularly important research direction is the distinction between technological feasibility at a given site and the feasible reconfiguration of a network under uncertain supply. Estimating both requires process-level information and a model of joint constraints. It can show whether investment in a local technology reduces a binding aggregate constraint or creates additional capacity at a nonbinding stage.
Conclusion
The production possibility frontier is the economic representation of maximal attainable output combinations within a specified production environment. Its geometry expresses the relationships among scarce resources, technologies, and alternative outputs. The marginal rate of transformation gives the local opportunity cost along the frontier; its numerical meaning depends on output units, the allocation point, and the set of binding constraints.
A movement along a frontier reflects the selection of a different production combination. A movement toward it can reflect improved utilization. A change in the frontier can arise from altered resource availability, accessible techniques, or institutional conditions. Technological change is rigorously identified only through a comparison holding the other relevant conditions fixed or explicitly accounting for their variation. The magnitude and direction of an expansion may differ across outputs, inputs, firms, and regions.
Convexity, divisibility, and competitive price support are substantive assumptions. The nonconvex example establishes that efficient output configurations need not be decentralizable through linear prices. Dynamic feasibility also depends on investment, adoption costs, uncertainty, and future opportunities. An expanded long-run frontier can be accompanied by a costly transition.
In empirical research, engineering benchmarks, peer-based efficiency analysis, stochastic frontier methods, and growth accounting answer distinct questions. Measured productivity growth may combine technological progress, resource reallocation, scale, quality, and unmeasured inputs. Trade and infrastructure change feasible configurations without necessarily changing the technical process of an individual producer. Environmental and service outputs demand careful attention to measurement and joint production.
The scientific value of frontier analysis lies in making opportunity and sacrifice explicit, identifying the constraints that govern production, and specifying the evidence needed to distinguish technological progress from other observable changes. Economic interpretation becomes more reliable when physical feasibility, market valuation, institutional access, and social evaluation are treated as separately examinable elements of one production problem.
References
Acemoglu, D. (2002). Directed technical change. The Review of Economic Studies, 69(4), 781–809. doi:10.1111/1467-937X.00226. Document
Aigner, D., Lovell, C. A. K., & Schmidt, P. (1977). Formulation and estimation of stochastic frontier production function models. Journal of Econometrics, 6(1), 21–37. doi:10.1016/0304-4076(77)90052-5. Document
Arrow, K. J., & Debreu, G. (1954). Existence of an equilibrium for a competitive economy. Econometrica, 22(3), 265–290. doi:10.2307/1907353. Document
Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research, 2(6), 429–444. doi:10.1016/0377-2217(78)90138-8. Document
Farrell, M. J. (1957). The measurement of productive efficiency. Journal of the Royal Statistical Society: Series A (General), 120(3), 253–281 (article; discussion follows). doi:10.2307/2343100. Document
International Renewable Energy Agency. (2025). Renewable power generation costs in 2024. Abu Dhabi: IRENA. Document
IPCC. (2022). Mitigation pathways compatible with long-term goals. In P. R. Shukla et al. (Eds.), Climate change 2022: Mitigation of climate change (Working Group III contribution to the Sixth Assessment Report, Chapter 3). Cambridge University Press. Document
Marshall, A. (1920). Principles of economics (8th ed., Book V, Chapter I). London: Macmillan. (Original work published 1890). Document
Mueller, N. D., Gerber, J. S., Johnston, M., Ray, D. K., Ramankutty, N., & Foley, J. A. (2012). Closing yield gaps through nutrient and water management. Nature, 490, 254–257. doi:10.1038/nature11420. Document
Mueller, N. D., Gerber, J. S., Johnston, M., Ray, D. K., Ramankutty, N., & Foley, J. A. (2013). Corrigendum: Closing yield gaps through nutrient and water management. Nature, 494, 390. doi:10.1038/nature11907. Document
OECD. (2001). Measuring productivity: OECD manual: Measurement of aggregate and industry-level productivity growth. Paris: OECD Publishing. doi:10.1787/9789264194519-en. Document
OpenStax. (2022). Principles of economics (3rd ed.), Section 2.2: The production possibilities frontier and social choices. Houston, TX: OpenStax. Document
Romer, P. M. (1990). Endogenous technological change. Journal of Political Economy, 98(5, Part 2), S71–S102. doi:10.1086/261725. Document
Solow, R. M. (1957). Technical change and the aggregate production function. The Review of Economics and Statistics, 39(3), 312–320. doi:10.2307/1926047. Document
ten Raa, T. (2008). Debreu’s coefficient of resource utilization, the Solow residual, and TFP: The connection by Leontief preferences. Journal of Productivity Analysis, 30, 191–199. doi:10.1007/s11123-008-0112-8. Document
U.S. Bureau of Labor Statistics. (n.d.). Handbook of methods: Major sector productivity and costs — concepts. Document
World Bank. (n.d.). World Bank Enterprise Surveys: Methodology. Document